FROM TRIANGULATED CATEGORIES TO CLUSTER ALGEBRAS

Description

Niveau: Supérieur, Doctorat, Bac+8
FROM TRIANGULATED CATEGORIES TO CLUSTER ALGEBRAS PHILIPPE CALDERO AND BERNHARD KELLER Abstract. The cluster category is a triangulated category introduced for its combinato- rial similarities with cluster algebras. We prove that a cluster algebra A of finite type can be realized as a Hall algebra, called exceptional Hall algebra, of the cluster category. This realization provides a natural basis for A. We prove new results and formulate conjectures on ‘good basis' properties, positivity, denominator theorems and toric degenerations. 1. Introduction Cluster algebras were introduced by S. Fomin and A. Zelevinsky [12]. They are subrings of the field Q(u1, . . . , um) of rational functions in m indeterminates, and defined via a set of generators constructed inductively. These generators are called cluster variables and are grouped into subsets of fixed finite cardinality called clusters. The induction process begins with a pair (x, B), called a seed, where x is an initial cluster and B is a rectangular matrix with integer coefficients. The first aim of the theory was to provide an algebraic framework for the study of total positivity and of Lusztig/Kashiwara's canonical bases of quantum groups. The first result is the Laurent phenomenon which asserts that the cluster variables, and thus the cluster algebra they generate, are contained in the Laurent polynomial ring Z[u±11 , . . .

• variables associated


• kq-modules


• quantum group


• auslander-reiten translation


• mod kq


• aim has


• positive root


• cluster variables


• indecomposable kq-modules

Sujets

• Quantum group
• Root system

Informations

• Publié par: mijec
• Publié le: 01 mai 2006
• Nombre de lectures: 38
• Langue: English

Extrait

FROM TRIANGULATED CATEGORIES TO CLUSTER ALGEBRAS

PHILIPPE CALDERO AND BERNHARD KELLER

Abstract.cluster category is a triangulated category introduced for its combinato-The rial similarities with cluster algebras. We prove that a cluster algebraAof finite type can be realized as a Hall algebra, called exceptional Hall algebra, of the cluster category. This realization provides a natural basis forA prove new results and formulate conjectures. We on ‘good basis’ properties, positivity, denominator theorems and toric degenerations.

1.Introduction

Cluster algebras were introduced by S. Fomin and A. Zelevinsky [12]. They are subrings of the fieldQ(u 1 , . . . , u m ) of rational functions inmindeterminates, and defined via a set of generators constructed inductively. These generators are calledcluster variablesand are grouped into subsets of fixed finite cardinality calledclusters induction process. The begins with a pair (x, B), called aseed, wherexis an initial cluster andBis a rectangular matrix with integer coefficients. The first aim of the theory was to provide an algebraic framework for the study of total positivity and of Lusztig/Kashiwara’s canonical bases of quantum groups. The first result is theLaurent phenomenonwhich asserts that the cluster variables, and thus the cluster algebra they generate, are contained in the Laurent polynomial ringZ[u 1 ±1 , . . . , u ± m 1 ]. Since its foundation, the theory of cluster algebras has witnessed intense activity, both in its own foundations and in its connections with other research areas. One important aim has been to prove, as in [1], [31], that many algebras encountered in the theory of reductive Lie groups have (at least conjecturally) a structure of cluster algebra with an explicit seed. On the other hand, a number of recent articles have been devoted to establishing links with subjects beyond Lie theory. These links mainly rely on the combinatorics by which cluster variables are grouped into clusters. Among the subjects concerned we find Poisson geometry[17],whereclustersareinterpretedintermsofintegrablesystems,Teichm¨uller theory [11], where clusters are viewed as systems of local coordinates, and tilting theory [5], [3], [15], where clusters are interpreted as sets of indecomposable factors of a tilting module. A cluster algebra is said to be offinite typeif the number of cluster variables is finite. In [13], S. Fomin and A. Zelevinsky classify the cluster algebras of finite type in terms of Dynkin diagrams. The cluster variables are then in bijection with thealmost positive roots of the corresponding root system,i.e.roots which are positive or opposite tothe simple roots. Note that this classification is analogous to P. Gabriel’s classification of representation-finite quivers; it is also analogous to the classification of finite-dimensional semisimple Lie algebras. In finite type, the combinatorics of the clusters are governed by generalized associahedra. The purpose of the decorated categories of [25] and, later, of the cluster categories of [5], [3], [8], is to offer a better understanding of these combinatorics. LetQbe a quiver whose underlying graph is a simply laced Dynkin diagram and let modkQbe the category of finite-dimensional representations ofQover a fieldk cluster. The

Date: version du 20/05/2006.

1

2 PHILIPPE CALDERO AND BERNHARD KELLER categoryCis the orbit category of the bounded derived categoryD b (modkQ) under the action of a certain automorphism. Thus, it only depends on the underlying Dynkin diagram and not on the orientation of the arrows in the quiverQ. The canonical automorphism is chosen so as to extend the bijection between indecomposablekQ-modules and positive roots to a bijection between indecomposable objects ofCand almost positive roots and, hence, cluster variables. By the results of [5], this bijection also induces a bijection between clusters and ‘tilting objects’ ofC ‘coincidences’ lead us to the following. These

Question 1.Can we realize a cluster algebra of finite type as a ‘Hall algebra’ of the corresponding categoryC?

A first result in this direction is the cluster variable formula of [7]. This formula gives an explicit expression for the cluster variable associated with a positive rootαcorresponding to an indecomposable moduleM α  exponents of the Laurent monomials of the cluster: The variableX α are provided by the homological form onmodkQand the coefficients are Euler characteristics of Grassmannians of submodules ofM α . In the present article, we use this formula to provide a more complete answer to Question 1 and to obtain structural results on the cluster algebra, some of which constitute positive answers to conjectures by Fomin and Zelevinsky. We first describe our structural results: *Canonical basis.We obtain aZ-basis of the cluster algebra labelled by the set of so-called exceptional objects of the categoryC. The results below point to an analogy of this basis with Lusztig/Kashiwara’s dual canonical bases of quantum groups. *Positivity conjecture.We prove that the Laurent expansion of a cluster variable has positive coefficients, when the seed is associated to an orientation of a given Dynkin diagram. Note that the proof of the positivity relies on the cluster variable formula and not on the multiplication formula described below. *Good basis property and Toric degenerations.We prove that these bases are compatible with ‘good filtrations’ of the cluster algebra. This provides toric degenerations of the spectrum of cluster algebras of finite type, in the spirit of [6]. *Denominator conjecture. the denom-The formula enables us to prove the following: inator of the cluster variable associated to the positive rootαseen as a rational function in its reduced form in theu i ’s is Q i u n i i whereα= P i n i α i is the decomposition ofαin the basis of simple roots. Note that this result can be found in [9]. It was first proved in [13] whenQhas an alternating orientation. Now, the main result of this article is the ‘cluster multiplication theorem’, Theorem 2. This result yields a more complete answer to Question 1. It provides a ‘Hall algebra type’ multiplication formula for the cluster algebra. The main part of the paper is devoted to the proof of this formula. A different proof has been recently obtained in [20] using the main result of [18]. Recall that by a result of [22], the categoryC cluster multiplicationis triangulated. The formula expresses the product of two cluster variables associated with objectsLandNof the cluster category in terms of Euler characteristics of varieties of triangles with end terms LandNleads to an expression of the structure constants use of the formula . Repeated of the cluster algebra in the basis provided by the exceptional objects in terms of Euler characteristics of varieties defined from the triangles of the categoryC the cluster. Thus, algebra becomes isomorphic to what we call the ‘exceptional Hall algebra’ ofC. This theorem can be compared with Peng and Xiao’s theorem [26], which realizes Kac- Moody Lie algebras as Hall algebras of a triangulated quotient of the derived category of a hereditary category. But a closer look reveals some differences. Indeed, we use the quotient

FROM TRIANGULATED CATEGORIES TO CLUSTER ALGEBRAS 3 of the set of trianglesW Y N,M of the form M→Y→N→M[1], by the automorphism groupAutYof the objectY, while Peng and Xiao use the quotient W N Y M, /Aut(M)×Aut(N). Hence, our approach is more a ‘dual Hall algebra’ approach as in Green’s quantum group realization [18]. Another difference is that the associativity of the multiplication is not proved a priori but results from the isomorphism with the cluster algebra. The paper is organized as follows: Generalities and auxiliary results on triangulated categories and, in particular, on the cluster category are given in Section 2 and in the appendix, where we prove the constructibility of the setsW MN, Y /Aut(Y) described above. Section 3 deals with the cluster multiplication formula. We first reduce the proof to the case where the objects involved are indecomposable. Then the indecomposable case is solved. Here, the homology functor fromCto the hereditary category of quiver repre- sentationsplaysanessentialrˆole.Itallowsustobypassthe‘triangulatedgeometry’ofC, which unfortunately is even out of reach of the methods of [33], [34], because the graded morphism spaces of the categoryCare not of finite total dimension. main ingredient The of the proof is the Calabi-Yau property of the cluster category, which asserts a bifunctorial duality betweenExt 1 (M, N) andExt 1 (N, M) for any objectsMandN. In Section 4, we use Lusztig’s positivity results for canonical bases [23], [24], to prove the positivity theorem. Then we obtain the denominator theorem. Section 5 deals with good bases for these cluster algebras. We provide a basis indexed by exceptional objects of the categoryC,i.e.objects without self-extensions. The cluster variable formula yields that this basis has a ‘Groebner basis’ behaviour and provides toric degenerations. The last part is concerned with conjectures for ‘non hereditary’ seeds of a cluster algebra of finite type. We formulate a generalization of the cluster variable formula, and other conjectures which would follow from it, such as results on positivity and simplicial fans. We close the article with a positivity conjecture for the multiplication rule of the exceptional Hall algebra.

Acknowledgments.The authors thank F. Chapoton, C. Geiss and A. Hubery for stimu- lating discussions. They are grateful to A. Zelevinsky for pointing out an error in an earlier version of this article.

2.The cluster category

2.1. Let Δ be a simply laced Dynkin diagram andQa quiver with underlying graph Δ. We denote the set of vertices ofQbyQ 0 and the set of arrows byQ 1 . Letkbe a field. We denote bykQthe path algebra ofQand bymodkQthe category of finitely generated rightkQ-modules. Fori∈Q 0 , we denote byP i the associated indecomposable projective kQ-module and byS i the associated simple module. Grothendieck group TheG 0 (modkQ) is free abelian on the classes [S i ],i∈Q 0 , and is thus isomorphic toZ n , wherenis the number of vertices ofQ any object. ForMinmodkQ, the dimension vector ofM, denoted by dim (M), is the class ofMinG 0 (modkQ). Recall that the categorymodkQis hereditary,i.e.we haveExt 2 (M, N) = 0 for any objectsM,NinmodkQ all. ForM,NinmodkQ, we put [M, N] 0 = dimHom(M, N),[M, N] 1 = dimExt 1 (M, N),hM, Ni= [M, N] 0 −[M, N] 1 . In the sequel, for any additive categoryF, we denote byind(F) the subcategory ofFformed by a system of representatives of the isomorphism classes of indecomposable objects inF.